Foundry Rolling Shutter

[Chris Millar]

http://www.thefoundry.co.uk/products/rollingshutter/

 

Anyone got an idea of how they remove rolling shutter ?

 

The algorithm in verbose pseudo code thanks! :blink:

 

I can see how in some situations after comparing frames and extracting the pan/tilt information (like match moving software does) the information for one frame can be extracted from another to give you back a reasonable image - albeit a little slit-scannish with all the temporal shuffling of lines. But some camera movements I would have thought would have no such past or present frames in which to extract the required lines and leave you with chunks of non-image to be filled with best guess pixels - bit of a mess yeh ?

 

And that will work with tilt/pan camera movement - but what about static shots where an object is moving ? I know match moving softwares (boujou et al) have algorithms are adjusted to account for this 'mode' - but I'm not sure if Rollingshutter can deal with this huh

 

Have they just used examples that work in their favour ?

 

 

interesting stuff

[Phil Rhodes]

It looks for motion within the frame and attempts to separate things into layers, or at least areas which are moving relative to one another. It then does its best to fill in the gaps using information from other frames. There are of course non-fixable situations where there is not enough information to do things properly, regardless of how well the algorithm works.

 

P

[Chris Millar]

Was going to ask why this isn't this happening in camera yet, but I figure that cheaper camera users don't care or know about the issue, and EX and so on users dont want the artefacts of when the algorithm fails ...

 

I wonder if there is a way you could have a hybrid system, where it knew it was going to fail (beyond some user configurable limit) and then leave the footage alone for the time being.. Then give you (or some software down the line) some sort of meta info of where and when it was doing its business so the original rolling frames could be extracted if need be (the meta info could also hold any 'unused/redundant' pixels, that may have been filled by repeats previously)...

 

its an interesting shuffling of temporal data but for a spatial effect - no reason the shuffling info can't be saved

 

market for this = me

:lol:

[Adrian Sierkowski]

It requires a fair bit of processing power, one would guess, which would kill your battery life, increase your costs, and create heat in camera causing issues....

Best guess, of course.